I'm not sure what you're going to do with the output, though. Using the same delimiter for multiple items does make things more ... entertaining. What you have to start with is something you can probably feed into DBD::CSV and do queries against. What you're transforming it into will only be parsable by custom code (possibly based on Text::CSV_XS).

If you want to reduce duplication yet still maintain some orthogonality, you could normalise your file into two files: one with ID/name/description, the second with ID and product. ID's would still be listed multiple times, but the name and description wouldn't be.

This would also lend itself to feeding into a more full-featured data store (such as Sqlite, mySQL, postgreSQL, or even DB2 or Oracle) at a later date for more advanced handling. (See my latest database-oriented project. The output is basically pure SQL calls against nearly normalised data.)